The swift evolution of automated and accurate image recognition and categorization systems is predominantly propelled by the rapid advancements in deep learning technology. In particular, the classification of handwritten characters has garnered increasing interest owing to its significant contributions to automation, particularly in the development of applications aimed at assisting individuals with visual impairments. Basically, the Devanagari script has 12 vowels, 36 consonant basic forms, 10 numeric characters, and a few special characters. The used dataset comprises 92000 distinct images of handwritten characters of 46 classes containing numerals and consonants of Devanagari script. The dataset is divided into 85% training and 15% test sets, with images stored in Portable Network Graphics (PNG) format at a resolution of 32x32 pixels. Various methods used to improve character recognition, including Deep Convolutional Neural Network (D-CNN) along with different deep learning algorithms such as LeNet, VGG and ResNet to train models. The studys results were presented and compared for each model and promising achieved 99.94 % training accuracy and 99.57% testing accuracy with the training loss of 0.020.

Deep Learning

Devanagari character recognition

Deep Convolution Neural Network (DCNN)

LeNet

VGG

ResNet